Hi All,
I want to run an ordered probit model with a binary endogeneous explanatory variable. I have 3 possible methods at my disposition:
1. Manually estimate the model using a 2-step method described as follows: Step 1: Run a probit for the endogeneous explanatory variable with an exclusion restriction included in the set of regressors, Step 2: Generate the generalized residuals from Step 1 and use as a regressor in the estimation of ordered probit,
2. Use the Stata package heckprobit,
3. Use the Stata package opsel.
I have attempted methods 1. and 2. I obtain almost similar results, but the coefficients are off by a few decimal points. Also for 2 variables the significance levels differ between the 2 methods (and I don't understand why so I am suspect of my results).
Finally, from my reading till now I understand that opsel employs a MLE procedure, which I understand is the most efficient method to estimate this model. However when I try to estimate the model using opsel I get the following error:
> factor variables and time-series operators not allowed
I would appreciate any help regarding:
1. Difference between the heckprobit and opsel commands. Is one superior to the other for my purpose? Any recommendation regarding study material is most welcome.
2. The error message that I receive when I use opsel.
3. I am most comfortable estimating the model by the 2 step method (since I can see exactly what is happening). Is it recommended to use a manual method even when Stata packages are available? Any caveats I should be wary of?
Thank you.
Aakanksha
0 Response to Ordered probit with binary endogeneous explanatory variable: heckprobit vs. opsel vs. generating generalized residuals manually
Post a Comment